07.01.2013 Views

Lecture Notes in Computer Science 3472

Lecture Notes in Computer Science 3472

Lecture Notes in Computer Science 3472

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

1 Hom<strong>in</strong>g and Synchroniz<strong>in</strong>g Sequences 31<br />

Length of Synchroniz<strong>in</strong>g Sequences and Čern´y’s Conjecture. Synchroniz<strong>in</strong>g<br />

sequences were <strong>in</strong>troduced a bit later by Čern´y[Čer64] and studied mostly<br />

<strong>in</strong>dependently from hom<strong>in</strong>g sequences, with some exceptions [Koh78, Rys83,<br />

LY96]. The focus has largely been on the worst-case length of sequences, except<br />

an article by Rystsov that classifies the complexity of several related problems<br />

[Rys83], the article by Eppste<strong>in</strong>, which <strong>in</strong>troduces the algorithm <strong>in</strong> Section 1.3.2<br />

[Epp90], and the survey by Lee and Yannakakis [LY96]. Čern´y [Čer64] showed<br />

an upper bound of 2n − n − 1 for the length of synchroniz<strong>in</strong>g sequences and<br />

conjectured that it can be improved to (n − 1) 2 , a conjecture that <strong>in</strong>spired much<br />

of the research <strong>in</strong> the area. The first polynomial bound was 1<br />

2n3 − 3<br />

2n2 + n +1<br />

due to Starke [Sta66], and as mentioned <strong>in</strong> Section 1.3.2 the best known bound is<br />

1<br />

6 (n3 −n) due to Klyachko, Rystsov and Spivak [KRS87]. Already Čern´y [Čer64]<br />

proved that there are automata that require synchroniz<strong>in</strong>g sequences of length<br />

at least (n − 1) 2 , so if the conjecture is true then it is optimal.<br />

Prov<strong>in</strong>g or disprov<strong>in</strong>g Čern´y’s conjecture is still an open problem, but it has<br />

been settled for several special cases: Eppste<strong>in</strong> [Epp90] proved it for monotonic<br />

automata, which arise <strong>in</strong> the orientation of parts that we saw <strong>in</strong> Example 1.3;<br />

Kari [Kar03] showed it for Eulerian mach<strong>in</strong>es (i.e., where each state has the<br />

same <strong>in</strong>- and out-degrees); P<strong>in</strong> [P<strong>in</strong>78b] showed it when n is prime and the<br />

mach<strong>in</strong>e is cyclic (mean<strong>in</strong>g that there is an <strong>in</strong>put letter a ∈ I such that the<br />

a-transitions form a cycle through all states); Čern´y, Pirická and Rosenauerová<br />

[ ČPR71] showed it when there are at most 5 states. Other classes of mach<strong>in</strong>es<br />

were studied by P<strong>in</strong> [P<strong>in</strong>78a], Imreh and Ste<strong>in</strong>by [IS95], Rystsov [Rys97], Bogdanović<br />

et al. [BIĆP99], Trakhtman [Tra02], Göhr<strong>in</strong>g [Göh98] and others. See<br />

also Trakhtman’s [Tra02] and Göhr<strong>in</strong>g’s [Göh98] articles for more references.<br />

Parallel Algorithms. In a series of articles, Ravikumar and Xiong study the<br />

problem of comput<strong>in</strong>g hom<strong>in</strong>g sequences on parallel computers. Ravikumar gives<br />

a determ<strong>in</strong>istic O( √ n log 2 n) time algorithm [Rav96], but it is reported not to<br />

be practical due to large communication costs. There is also a randomized algorithm<br />

requir<strong>in</strong>g only O(log 2 n) time but O(n 7 ) processors [RX96]. Although not<br />

practical, this is important as it implies that the problem belongs to the complexity<br />

class RNC. The same authors also <strong>in</strong>troduced and implemented a practical<br />

randomized parallel algorithm requir<strong>in</strong>g time essentially O(n 3 /k), where the<br />

number k of processors can be specified [RX97]. It is an open problem whether<br />

there are parallel algorithms for the synchroniz<strong>in</strong>g sequence problem, but <strong>in</strong> the<br />

special case of monotonic automata, Eppste<strong>in</strong> [Epp90] gives a randomized parallel<br />

algorithm. See also Ravikumar’s survey of parallel algorithms for automata<br />

problems [Rav98].<br />

Nondeterm<strong>in</strong>istic and Probabilistic Automata. The hom<strong>in</strong>g and synchroniz<strong>in</strong>g<br />

sequence problems become much harder for some generalizations of Mealy<br />

mach<strong>in</strong>es. As shown <strong>in</strong> Exercise 1.8, they are PSPACE-complete for nondeterm<strong>in</strong>istic<br />

automata, where δ(s, a) is a subset of S. This was noted by Rystsov<br />

[Rys92] as a consequence of the PSPACE-completeness theorem <strong>in</strong> another of

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!